This repository is an official implementation of PrevPredMap.
This project introduces a novel temporal modeling framework, named PrevPredMap, which pioneers the use of previous predictions for the construction of online vectorized HD maps. Two essential modules for PrevPredMap are meticulously crafted: the previous-predictions-based query generator and the dynamic-position-query decoder. These modules not only enable PrevPredMap to effectively encode and utilize previous predictions but also ensure robust performance across both single-frame and temporal modes.
This image displays three common scenarios in autonomous driving, where the camera's field of view is partially obstructed by a large truck, plastic water-filled barriers, and roadside railings, respectively.
Method | Backbone | Epoch | Divider | Ped. Crossing | Boundary | mAP | Config | Download |
---|---|---|---|---|---|---|---|---|
PrevPredMap | R50 | 24 | 65.9 | 65.8 | 67.8 | 66.5 | config | model & log |
PrevPredMap | R50 | 110 | 70.1 | 71.2 | 72.8 | 71.3 | config | model & log |
These settings are mostly the same as MapTRv2
PrevPredMap is based on MapTRv2 and mmdetection3d. It is also greatly inspired by the following outstanding contributions to the open-source community: GroupDETR, BEVFormer, GKT.
If you find PrevPredMap is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.
@article{peng2024prevpredmap,
title={PrevPredMap: Exploring Temporal Modeling with Previous Predictions for Online Vectorized HD Map Construction},
author={Peng, Nan and Zhou, Xun and Wang, Mingming and Yang, Xiaojun and Chen, Songming and Chen, Guisong},
journal={arXiv preprint arXiv:2407.17378},
year={2024}
}